How Amazon and Walmart may be disrupted by ChatGPT and Perplexity

Now, he’s onto another e-commerce bet for the AI age after spending much of the past decade launching and running an on-demand car service business. His new company, ReFiBuy (an abbreviation of research, find, buy),  aims to build software that helps consumer brands and retailers navigate a new online world where AI shopping agents make purchases as frequently as human customers do. 

Fortune recently caught up with Wingo to discuss how generative AI and AI agents are disrupting online shopping, and how e-commerce powerhouses may be impacted by this moment. Not surprisingly, as someone attempting to build a big new business in the space, Wingo is perhaps more bullish than many about how soon and severely new AI shopping tools may disrupt existing e-commerce giants and legacy retailers alike. The interview has been edited for length and clarity.

Fortune: What did you see going on in e-commerce that made you want to get back into it? 

Scott Wingo: The LLM companies [like ChatGPT, or Perplexity] are getting consumer attention, so they’re getting consumer distribution. And when you get that consumer distribution, you have to figure out, how do I monetize it? It’s become pretty apparent to me – especially when Perplexity launched their first [shopping feature] – that this was going to be the next big wave of e-commerce, and I couldn’t miss that. I feel like my team and I have a unique view being subject matter experts and seeing how e-commerce has been done so far – the good and the bad of it.

What’s the “bad” you are referencing?

If you look at e-commerce in the U.S., the growth level has really slowed down, right? Depending on which numbers you look at and what you count, we’ve settled in somewhere between 15% and 20% of overall retail sales comes from e-commerce, and that’s been a pretty steady number.

But if you look at other countries, that number feels wrong. You look at Europe and that number is way higher. And Asia is way off the charts higher. So why is that? And I think it’s because our user experiences have really stalemated or gone stagnant. You’re a big Amazon follower, as am I, but something has happened where they have totally stopped innovating on retail and have just started cash-cowing it.

Since you mentioned Amazon, how do you see it being affected in this new world?

So they’ve got the retail business, a [third-party seller] marketplace, and ads. 

The ad layer in my mind doesn’t add value to the consumer experience. So imagine a world where someone builds [an AI experience] on top of Amazon and no one’s using the Amazon front end anymore – they’re starting with an AI chat or agent. Amazon’s just a backend now, so humans aren’t seeing those product ads on Amazon. So there goes $60 billion of profit margin basically. That’s problem No. 1.

Problem No. 2 is now you’re just the backend for a different front end [like ChatGPT or Perplexity], and at some point you’re gonna have to pay a tax [to the AI company], because you could get switched out. So Amazon hates paying a tax. They are the tax collector, right? So there’s gonna be a fight for who’s the tax collector. So that’s gonna be interesting. 

Then I think the third thing is, if I’m ChatGPT or [OpenAI CEO] Sam Altman, what if I only charged a 5% take rate to sellers to list their products (vs 12% to 15% for many categories on Amazon)? “You can keep your products listed on Amazon, but we’d also love them to be listed with us too. And since we’re cheaper, would you consider offering a lower price?” A brand like Dyson [might] not take that deal, but the small business seller is maybe going to take that offer. 

So there’s a way you could really attack Amazon by getting rid of the ad business quickly, which would be pretty detrimental. And then I think you could chew away at a big chunk of the marketplace business. And those are the two – AWS aside – profit centers of the company.

When you talk about someone building on top of Amazon, you’re talking about AI search engines or shopping agents that become the starting point for more customers for online shopping. How will you or I search for products or shop differently through AI shopping agents than we do today?

So first, it’ll be multimedia – it could be voice, typed, or whatever you want it to be. I think we’ll gravitate to voice because it’s gonna get really good. 

You’ll have this conversation with what will be your personal assistant, and it’s gonna be doing everything for you. So you’ll be having a conversation with it about your calendar, and it’s going to tell you, “Remember you got this dinner tonight, and Bob’s going to be sitting to your left, and he’s the CTO of this company and remember you have this story coming up about his company.” It’s basically going to become an advanced assistant for you. 

And then it’ll say, “By the way, can I have a quick conversation with you about the things you need to replenish?” And you’ll say, “Sure.” It’s like, “Well, I got the toilet paper coming in on this day, the milk on this day, do you approve?” And you’ll say, “Sure.” Or you’ll just trust it so much it’ll just order it. That’s how the basics will happen. It’ll just be a quick conversation or you just set it to autopilot.

And on the back end, who’s involved there? 

Maybe it is Amazon. Maybe it’s Walmart. The agent or assistant will decide that for you. You’ll give it parameters, maybe on value or convenience. So the value would be, “I don’t care where the milk comes from, as long as it’s under X bucks a gallon and it’s 2%.”

Then it gets really weird, because now the agents could run little auctions and be talking to each other and saying, “I’ve got this big replenish order, who wants it?” 

On the convenience side, you’re buying something new, and it’s a considered purchase – say, you want to spend 150 bucks on a pickleball paddle. And it will know the right questions to ask you and suggest the attributes you need to find. And maybe you’re like, “I need it in two days, and I need to pay with [the Buy Now, Pay Later option] Affirm.” 

How much can Amazon’s new AI features like its Rufus AI shopping assistant help its battle?

They’re good efforts, but I believe ChatGPT is going to be better. Part of it is that Rufus can’t be so good that you don’t need [Amazon’s product search] advertising. For Rufus to improve dramatically, it almost has to replace the existing search experience on Amazon. And that’s a chasm Amazon’s not going to cross because that would kill $60 billion of advertising revenue that’s essentially pure profit margin.

Do you get any sense from the conversations you’re having that Amazon or Walmart are aware of the size of the threat, at least as you see it?

I haven’t had conversations with Amazon, but Walmart people have very publicly said we’re going to take a very open approach to shopping agents and welcome them. In a way, that’s smart because they’re a challenger to Amazon, and I don’t think they have anything to lose. This is a chance for them to level the playing field and kind of catch up to Amazon. They’re okay if they’re not the front end for all these transactions, because they also have the stores and a lot of their online orders get picked up in the store, so at least they’ll have consumer interaction at that layer.

What are your top one or two messages to retailers or brands on how their operations need to change, or what they need to be doing differently than they’re doing now?

So the honest answer is, we think this is just the start, and it’s going to be bad, and what you have to do is embrace these agentic systems, and your data is not ready for it. 

There’s a couple pools of data that we have in the retail world, and they’re all not really ready for agents. So there’s no standard for product catalogs. This is what I lived at ChannelAdvisor – everyone would send us their product catalog, and they were a hot mess. 

Pick any category and everyone has different attributes (characteristics or details describing a piece of merchandise) for every product, and they use different terminology. Let’s say a gold standard for a GoPro 10 had 200 attributes. Any given retailer may only have 20 of those. Maybe if you’re Amazon and Walmart, you have 50 of those attributes. 

But the agentic engines need more content and context. The average product search may include four terms, but AI queries or chats are like 15 to 20 words. To bridge these things, we want to help retailers look at their data, evaluate it, and kind of say, here’s where you’re doing okay, and here’s where you’re not, here’s where you’re prepared, and here’s where you’re super unprepared. And then we’ll have capabilities to go and help. 

How fast is this all going to happen in your view?

I started thinking about this idea and talking to people in Q4 2024, but over the last 30 days, it’s gone from kind of interesting for really large multi-billion-dollar retailers to “I am getting pressure from all sides to figure this out.” So there’s board level pressure. There’s a lot of that. But what they’re really seeing is the distribution of inbound traffic is changing very dramatically, very quickly. In May, it started in e-commerce and people are behind the scenes, hair on fire, freaking out about it.